PAWLIN Technologies” LLC developed algorithms for recognizing faces by identifying facial features, collecting reference points and objects from the image of subject’s face. Our algorithms for face recognition are based on the most advanced deep learning neural networks.“
Nowadays, on training sample, using combination of algorithms and large amount of data, we were able to achieve the 98.2% accuracy during internal Labeled Faces in the Wild testing.
However, it is worth to be considered that such criteria as location and camera resolution, lighting and movement speed characteristics, can affect the accuracy.
Our algorithms allow to search and identify faces in both photos and video streams.
- development flexibility, customization for solving special problems and conditions
- integration into the customer’s system
One on the advantages of our software is cross-platform feature. Moreover, you can process and get analytics on your server or in the cloud with only one condition: the presence of Internet/Ethernet channel. You can download your data in PDF and Excel.
Example of data collection on the server
The software uses the face and the body as initial data for analysis and for keeping the uniqueness of processed objects. To make the recognition successful the minimum resolution for embedded wired camera should be HD or FHD.
The maximum distance from the face depends on the lens (~ 100 m), the permissible face inclination/rotation angle without reducing the accuracy is set at 25-30 degrees.
The size of face identifier is only 1kb, and the minimum resolution of the image, suitable for recognition without reducing the accuracy is 30 pixels between the eyes.
We continue improving and expanding the functionality of our solution. In the near future we plan to add recognition of human emotions, search and recognition of logos (brands) in the image (video stream).
We use Face Recognition algorithms both in our own and in third-party software and hardware solutions. If necessary, integration into the customer’s system is possible.
To test our development, you can use REST API to get access to server.